K5 raw is 14 bit

Ronald E. Chambers wrote:

The K5 raw output is 14 bits as compared to the 12 of the K7. Seems like the DR should be improved some with this.

RON C

Maybe... but in and of itself, that doesn't matter.

There's some number of photons the well can hold, and there's a limit that gets lost in read noise/shot noise/etc. This is the "analog" dynamic range of the photosite.

When I do my analog to digital conversion, I chop that up into a bunch of different levels, and assign the resulting photon count (charge of the photosite) to one of those bins.

Going from 12 to 14 bits just means I have 4 times as many bins that are 1/4 the size (for a given sensor) - it doesn't mean that the analog range of the sensor got larger.

With 14 bits, I can assign an intensity much more precisely within my range, but my range of sensitivity may or may not have changed.

Having said that, one of the reasons a company might choose to go to 14 bits is a larger analog range. (another is marketing, yet another is the IC package provided on the chip...)
Walter

etoastw wrote:

The K10D was 14bit raw, IIRC. However, Pentax went back to 12bit raw for the K20D and K7. Don't know why.

K10D was 12 bit as well -- Pentax claimed the processing was 22-bit or something like that, but that didn't seem to have much of an impact on anything.

not necessarily true... k10, at the time, had the largest DR by RAW analysis.. problem was banding w/ some sensors and it was nosier at higher iso's....
Anyways.. go to bottom.. at iso 100 K10 had excellent DR
http://daystarvisions.com/Docs/Rvws/K20D/pg3.html

As a side note Pentax software outputs fake 16bit TIFFS (actually 8) so some is lost anyways.....

This reference might make things a bit clearer for you. 14 bits do matter on larger pixel cameras. The use of high quality amplifiers also matters. Have a look at the section on DR particularly the graph and its explanation at 5a.

See

http://www.clarkvision.com/imagedetail/digital.sensor.performance.summary/index.html

That the 300s is not better than the D90 is proof that Nikon cut some corners somewhere, not that the theory is faulty.

rhlpetrus wrote:

From D90 and D300s comparison at DxO Mark it seems that for APS-C sensors, no extra DR is gained, maybe just some extra tone gradation.

Considering the impact that going to 14 bit has made on the depth of the buffer, I would hope that there is in fact some increase in DR.

Engineering is all about selecting the best compromise to match your product requirement. It would make no sense to reduce the number of frames that the buffer is capable of without actually gaining some performance in another area.

It would make no sense to reduce the number of frames that the buffer is capable of without actually gaining some performance in another area.

don't forget money as an area
--
Mike from Canada

'I like to think so far outside the box that it would require a telephoto lens just to see the box!' ~ 'My Quote :)'

http://www.airliners.net/search/photo.search?sort_order=views%20DESC&first_this_page=0&page_limit=180&&emailsearch=mighty_mike88%40hotmail.com&thumbnails=

saghost wrote:

Ronald E. Chambers wrote:

The K5 raw output is 14 bits as compared to the 12 of the K7. Seems like the DR should be improved some with this.

There's some number of photons the well can hold, and there's a limit that gets lost in read noise/shot noise/etc. This is the "analog" dynamic range of the photosite.

Ron and Walter, yes, there is a limit to the number of electrons the photosite "well" can hold, which from the Sony A55 that likely uses the same Sony sensor, is about 16,000 electrons at ISO 200 clipping limit, which is not much different than the older 12 MP sensor of the K-x on a per photosite basis ; however, the K-5 has an extra about 35% photosites in about the same sensor area. This affects the apparent noise vs. detail as to highlight detail graduations in the bright image tones at 100% zoom views due to "shot noise", which is the "physics" noise that arises from the statistical variation in the arrival of a finite number of photons. "Shot noise" affects dark shadow detail very little as the limit of this noise is zero when there are zero photons at absolute dark. The limit on the maximum number of photons captured per photosite at the clipping level affects the maximum ISO sensitivity gain that can be used for full size 100% zoom viewing as to the bright tones in that when there are less than about 400 electrons in the well, as would happen at about ISO 80,000 with this sensor, either one must reduce the detail level with Noise Reduction (NR) or reduce the maximum viewing size or increase viewing distance in order for the images to appear of acceptable quality.

As to the dark shadow detail limit, that is placed by the black read noise, which I can see from raws available from the Sony A55 are at about the same level as that of the K-x. Thus, I expect that the real Dynamic Range (DR) for the sensor from the brightest clipping level to the "noise floor" limit will be about the same as that of the K-x or about 12.5 stops at lowest ISO.

When I do my analog to digital conversion, I chop that up into a bunch of different levels, and assign the resulting photon count (charge of the photosite) to one of those bins.

Going from 12 to 14 bits just means I have 4 times as many bins that are 1/4 the size (for a given sensor) - it doesn't mean that the analog range of the sensor got larger.

With 14 bits, I can assign an intensity much more precisely within my range, but my range of sensitivity may or may not have changed.

Even if the raw black read noise was incredibly low at lowest ISO at about 1/8192 of the bright clipping level (13 stops of DR), even a signal at 10 stops down with a full well capacity 32,000 electrons at 10 stops below the bright clipping limit would have "shot noise" of about 1/5657 for a combined noise of 1/4655, meaning that all those extra "graduations" would have an almost imperceptible effect as they would just be "buried in the noise", with even more "shot noise" at brighter levels. The intensities that are being assigned with the extra bit depth is really just better defining the noise levels.

Having said that, one of the reasons a company might choose to go to 14 bits is a larger analog range. (another is marketing, yet another is the IC package provided on the chip...)

MightyMike suggested another alternative that might just make sense: Previous Pentax cameras have been criticized at DPR for not having as much raw "highlight headroom" as competing models. What if the extra two bits were at the top bright end of the range and extended the raw format's capacity to capture normally overexposed highlights without clipping, where they could be recovered by judicious application of tone curves and other raw processing? Would that not make the camera score very highly? Another benefit of doing this is that compressed raw file sizes would not likely increase that much in that these top levels would only generally be used by the small percentage of bright normally overexposed specular highlights, whereas using them as the least significant bits just generally captures high levels of random noise which is hard to compress and would make the files about four MBytes larger on average.

Another possibility is that it is further confusion on Pentax's marketing department in that the raw readings may be delivered as 14 bits but reduced to 12 bits for raw image file output just as the K20D, K200D, K-m/K2000D, and K-7 did.

In short from the above, it is very unlikely that 14 bits used other than as MightyMike suggested in extending the clipping limit would give any larger a real analogue range.

Yes, many companies (especially Canon) have gone to 14 bits for largely marketing reasons, and I sincerely hope that Pentax is not one of them, although I could accept using 14 bits to extend the highlight capture/roll-off range as described above.

If we are to believe the specifications we have been given, I doubt that the limit is the imaging engine used as it seems it is still Prime II just as for the K-7, K-x, and K-r, all of which use 12 bit processing although there is a 14 bit Analogue to Digital Converter (ADC) available. Although from the Pentax record for this K-5's release, that possibly should have read PRIME III!

Regards, GordonBGood

Richard Murdey wrote:

As others already pointed out, the sensor characteristics are not improved. 14 bit gives you is some additional precision that is most noticeable on flat, gradually changing tones like a blue sky. In the RAW data at least, there are now many more shades of blue.

Richard, even with the very best sensor with the lowest black read noise coupled with a zero noise conversion, the only place one would actually be able to see the difference due to very high bit depths would be in the very deep deep shadows. For instance, if we had a large sensor with an electron well depth of 100,000 electrons per photosite and a dark read noise of two electrons, we would have a Dynamic Range (DR) of 50,000 : 1 and could justify even a 16-bit conversion. However, we would only enjoy that low noise up to where the shot noise was very low, so that by the time the Signal to Noise Ratio (SNR) got up to about 4,000 : 1 at the 625 electron level or 7.3 stops below the full clipping level, the "shot noise" would be large enough to completely limit the use of that greater bit depth. Blue sky is much brighter than this, and only a few stops below the clipping level for normal exposures, and 14 bits would not make one little bit of difference in the amount of noise in these blue skies, as all of those extra fine shades will just be blurred together by the noise even if we were actually able to pick them out.

Worth remembering that any output you are likely to be looking at is no better than 8bit, it all gets downsampled in the end.

However, those eight non-linear bits are quite a bit different than the 12 or 14 linear bits in the raw image file, as the Tone Response Curve (TRC) applied in the conversion to viewable eight bit images actually compresses the bright levels and expands the dark levels in a way that more-or-less matches our human eye's (and monitor's) response to luminance levels in that we aren't actually able to perceive all of the 256 levels given by eight bits when our eyes are fully adjusted to the full range from full bright to maximum dark , as we can only actually perceive about 100 discrete steps between those levels.

Generally, we need at least about at least 10 "good" linear raw bits in order to convert to a normal eight bit JPEG with reasonable image quality.

Regards, GordonBGood

PhotoHawk wrote:

This reference might make things a bit clearer for you. 14 bits do matter on larger pixel cameras. The use of high quality amplifiers also matters. Have a look at the section on DR particularly the graph and its explanation at 5a.

See

http://www.clarkvision.com/imagedetail/digital.sensor.performance.summary/index.html

No, the 14 bits don't really matter for any current camera, even the Canon 1D Mark IV and the Canon 5D Mark II, as those high 15 stop Dynamic Range (DR) figures are only extrapolations for the actual sensor dark read noise taken from the high ISO response; at low ISO's these Canon models have too much dark read noise injected by their conversion electronics to allow such high DR results.

That the 300s is not better than the D90 is proof that Nikon cut some corners somewhere, not that the theory is faulty.

No, the reason that the Nikon D300s is not better at DR than the Nikon D90 even though the D300s can capture 14 bit raw image files is that the sensor dark read noise is not smaller than the least significant bits of a 12-bit range at the ISO sensitivity gain levels used for the D300 and D300s. To further confuse the issue, the D90 (and D5000) use some "sneaky" form of chroma Noise Reduction (NR) to make the noise appear to be lower even though this is at the cost of slightly reducing detail retention, especially in the deep shadows.

Regards, GordonBGood

John Cafarella wrote:

Considering the impact that going to 14 bit has made on the depth of the buffer, I would hope that there is in fact some increase in DR.

Engineering is all about selecting the best compromise to match your product requirement. It would make no sense to reduce the number of frames that the buffer is capable of without actually gaining some performance in another area.

John, that's the $100 question, why is the specified buffer depth so much less, especially if it turns out not to be a misprint in the specification?

I can't see that Pentax would actually reduce the size in MBytes of the buffer between the K-7 and the new top level K-5, although they may not have been able to increase the amount of buffer memory due to not redesigning the circuit boards to any large amount and not wanting to reduce battery life. Let's say that the K-7 had about 256 MBytes of buffer, which makes sense at about five frames per second (fps) in that about three frames would be flushed to flash memory in the about three seconds of continuous shooting time, which would then require space for about 14 raw files in the buffer or about 18.3 MBytes per raw file, which is just about what they are for the K-7 at low ISO and moderate detail level .

My theory about why the eight raw buffer limit exists (also with an accordingly low JPEG buffer limit) at maximum continuous rate is Pentax's typical conservative specification policy and the existence of these extremely high maximum ISO's in extended ISO mode . To me, it makes sense that the actual raw buffer limit is just proportionally less than that of the K-7 considering the larger pixel dimensions and the faster continuous shooting speed, such that 15 raw captures would seem to be about right for exactly the same shooting environment and ISO sensitivity settings . However, there very well might be only eight raw captures possible for worst case raw files sizes where the maximum raw file size at ISO 51,200 could be as large as 35 MBytes or so.

Thus, there is no discrepancy between the performance of the K-7 and this new K-5, and only confusion about the conditions under which that limit applies .

Regards, GordonBGood